Launching today
Clopen
All-in-one web workspace for Claude Code & OpenCode
3 followers
All-in-one web workspace for Claude Code & OpenCode
3 followers
A modern web interface for AI development with Claude Code & OpenCode. Manage multiple accounts and projects seamlessly. Includes a terminal, Monaco editor, a built-in browser where AI tests its own UI, git-like checkpoints with branching, and access from any device.
Vibe coding is powerful, but the workflow around it is still a mess.
You're switching windows just to check if an AI finished. You're logging in and out of Claude Code every time you switch clients. You're manually testing UI your AI just wrote. And when the AI breaks something, you're hoping git can save you.
We built Clopen to fix all of that. It's a local web workspace built specifically for AI-assisted development.
Key features:
🛠️ Everything in One Tab: Terminal, Monaco editor, file explorer, git controls, AI chat, and a live browser preview, all in a single browser tab. No more hunting for the right window.
🔔 Multi-Project Awareness: Run multiple AI agents across different projects simultaneously. Get bell notifications and unread indicators when any session finishes. Switch projects in one click, and background processes keep running while you're away.
🤖 Multi-Account & Engine: Save all your Claude Code accounts in one place and assign them per project. Switch between Claude Code and OpenCode per session instantly, no terminal, no config files.
⏪ Git-Like AI Checkpoints: Every chat is auto-saved as a checkpoint. Branch off and explore multiple directions at once. Restore any state completely without ever touching your real git history.
🌐 AI-Powered Browser Testing: A real browser lives inside the tab, right next to the chat. The AI renders its own UI, clicks through it, finds bugs, and fixes them without you touching the browser.
📱 Access from Anywhere: Built-in Public Tunnel lets you open your full workspace on any device. Secured with authentication so only you can access it.
Clopen runs fully locally and is open-source. Would love to hear your feedback and how you're using AI in your workflow! 🚀
One thing that kept happening: I'd be working on three projects at once, each open in its own Cursor window. After a while I'd genuinely lose track of which window was Project A and which was Project B. I'd click into the wrong one, realize it, switch again. Then find out an AI session had finished twenty minutes ago and had just been sitting there waiting. Nobody told me. I just happened to check.
Running multiple projects shouldn't mean babysitting every window. That's what pushed us to build the notification system and put everything in one tab.
I was building a dashboard and gave the AI three prompts back to back.
Chat A: "Build a dashboard. Stats widget on top, activity feed below." Looked great.
Chat B: "Connect the stats widget to the API. Add a date range filter." Still solid.
Chat C: "Add real-time updates to the activity feed via WebSocket." this broke the date range filter from Chat B. The WebSocket cycle was conflicting with the API fetch.
I tried describing the problem. Something else broke. Three chats deep, untangling something that was working fine two messages ago.
Git wasn't the answer. The AI had touched too many files, and reverting everything meant losing Chat B's work too.
So I restored to Chat B and tried a different direction.
Chat B2: "Add auto-refresh every 30 seconds instead." Date range filter still works. Everything holds.
But the WebSocket path didn't disappear. It became its own branch. So I went back to it and tried again.
Chat C2: "Rewrite the WebSocket handler to run separately from the data layer." Fixed.
Now I had two working solutions sitting side by side:
Branch 1: A → B → C → C2 (WebSocket, fixed)
Branch 2: A → B → B2 (polling, working)
Both accessible. I could compare them and decide which one to ship.
Claude Code's terminal only holds one active account at a time. That's just how it works.
If you're handling multiple clients, each with their own Claude account, every project switch means going to the terminal, logging out, authenticating again, and waiting. Every single time.
It's not dramatic. It's just five extra steps you do all day, every day. And every time you do it, you lose the thread of what you were working on.
We just moved all of that into a UI. Save your accounts once, assign one per project, and switching is a single click. The terminal never gets involved again.
The browser thing came from a genuinely frustrating afternoon.
I asked the AI to build a signup form. It did, quickly. But when I opened the browser to check it, the button was misaligned. I described it. Fixed. Then the label overlapped something. Described it. Fixed. Then on a narrower screen the whole layout broke. Described it again.
The AI was writing fast. I was the bottleneck, manually checking and translating what I saw into words every single round.
The AI has no idea what it's actually building looks like. It only knows what you tell it. So we gave it a browser panel inside the tab, right next to the chat. Now it opens that browser itself, clicks through what it built, finds what's wrong, and fixes it. I just watch.